基于A2范数的加权低秩子空间聚类

子空间聚类在运动分割、人脸聚类上得了广泛的应用，并且取得很好的聚类效果.针对稀疏子空间聚类和最小二乘回归子空间聚类求得的表示系数存在类内过于稀疏和类间过于稠密的问题，本文利用l₂范数，提出一种基于欧氏距离的且具有组效应的加权低秩子空间聚类算法，此算法通过基于欧氏距离的加权方式，使得最终的表示系数在保证同一子空间数据点联系的同时，减小不同子空间数据点之间的联系.利用此表示系数建立相似矩阵J，将J应用到谱聚类得到聚类结果.实验结果表明，与当前流行的算法比较，本算法取得了较好的聚类效果.

Weighted Low Rank Subspace Clustering Based on A2 Norm

Subspace clustering has been widely applied to motion segmentation, face clustering and has achieved good performance. In order to solve the problem of over-sparsity for within-class coefficients and over-density for between-class coefficients solved by SSC and LSR, this paper proposes a new subspace clustering based on Euclidean distance using l₂ norm. By the weighted method based on Euclidean distance, the coefficient representation obtainted by this algorithm remains the connections of the data points which are from the same subspace. Meanwhile, it can eliminate the connections between clusters. We can get the clusters by using the spectral clustering with the similarity matrix which is constructed by this coefficient representation. The results of experiments indicate our method improves the accuracy of clustering.